Marker

Marker converts PDFs, Word, PowerPoint, and images into clean Markdown, JSON, or HTML with high accuracy. Developers use it to prepare documents for AI pipelines and RAG. Note: the code is open-source but the models carry usage restrictions.

Marker is a document-conversion pipeline from Datalab that turns PDFs, images, and Office files (DOCX, PPTX, XLSX), plus HTML and EPUB, into clean Markdown, JSON, HTML, or RAG-ready chunks. It reads a document's embedded text, detects layout, formats tables, equations, and code blocks, strips page headers and footers, and extracts images. It only calls its vision model where the text layer is bad or scanned, which keeps it fast. An optional hybrid mode passes pages to an LLM (Gemini, Claude, OpenAI-compatible, Ollama, and others) for higher accuracy. It runs on GPU, CPU, or Apple Silicon and offers a CLI, a Python API, a small FastAPI server, and a Streamlit app. Teams building retrieval pipelines and document-processing workflows use it to get structured text out of messy real-world files. Note the licensing split: the code is Apache-2.0, but the model weights use a modified OpenRAIL-M license that is free for research, personal use, and companies under $5M in funding/revenue, with paid licensing required above that.